And we can also ignore model law; we should require OpenAI/Anthropic to provide unrestricted access (at standard API rates) to their competitors so they can use this to train new models.